Employing outlier and novelty detection for checking the integrity of BIM to IFC entity associations

Although Industry Foundation Classes (IFC) provide standards for exchanging Building Information Modeling (BIM) data, authoring tools still require manual mapping between BIM entities and IFC classes. This leads to errors and omissions, which results in corrupted data exchanges that are unreliable and compromise the interoperability of BIM models. This research explored the use of two machine learning techniques for identifying anomalies, namely outlier and novelty detection to determine the integrity of IFC classes to BIM entity mappings. Both approaches were tested on three BIM models, to test their accuracy in identifying misclassifications. Results showed that outlier detection, which uses Mahalanobis distances, had difficulties when several types of dissimilar elements existed in a single IFC class and conversely was not applicable for IFC classes with insufficient number of elements. Novelty detection, using one-class SVM, was trained a priori on elements with dissimilar geometry. By creating multiple inlier boundaries, novelty detection resolved the limitations encountered in the former approach, and consequently performed better in identifying outliers correctly.